Neighbor embedding is a technique used for determining predictions for residual encoding of image blocks and for reconstructing residual encoded image blocks.
In neighbor embedding reference blocks are determined for being combined for prediction or approximation of a block, determination of reference blocks being based on resemblance of spatial neighbourhoods of the reference blocks with a spatial neighborhood of the block. Resemblance between two blocks means that their semantic contents are similar.
This approximation or prediction can take place the same way at encoder side and at decoder side and only a residual of the prediction is conveyed or transmitted from the encoder to decoder.